RH v8.0 Performance

[Rick Stevens]

Karl Pearson wrote:
> On Thu, June 3, 2004 10:34 am, Rick Stevens said:
> 
>>Karl Pearson wrote:
>>
>>Have you done a "top" to see which processes are sucking up the CPU?
>>Those are the ones to tweak.
>>
>>Note that spamassassin can be a real CPU hog (perl often is)--expecially
>>if you have large corpuses to go through for spam classification.  I'm
>>not saying that's the culprit, but our systems use bogofilter (written
>>in C).
> 
> 
> I've run top for 'minutes' at a time and see that httpd and spamd are two
> of the most likely culprits, thus my attempt at increasing httpd
> performance before posting. When httpd is at the top, it's using only
> about 5% CPU, whereas when spamd is at the top of the list, it's using
> 99.9%. Same for clamd, but it's not at the top very often.

As I said, spamd can suck serious CPU due to the nature of Perl.  It's
not really intended for use on a mail gateway with a lot of traffic.
It's more of a single user thing.  Remember that when Perl starts up
it has to load a gazillion modules and libraries which slow down its
startup and eat lots of CPU and disk I/O.  You could fire up a system
monitor program such as gkrellm or gnome-system-monitor and see where
it's getting stuck.

I suppose you could renice spamd to give it a lower priority, but I
don't know how much that'd buy you.  It depends on where it's really
spending its time--firing up perl or running through the corpi.

For mail gateway use, I highly recommend bogofilter.  The whole thing is
written in C and is much more efficient.  Not as easy to tweak, but
works well and easy to integrate into procmail recipes.
----------------------------------------------------------------------
- Rick Stevens, Senior Systems Engineer     rstevens at vitalstream.com -
- VitalStream, Inc.                       http://www.vitalstream.com -
-                                                                    -
- Linux is like a wigwam...no windows, no gates...and apache inside! -
----------------------------------------------------------------------